May 28, 2026

In this edition

AI Assist & Tooling

Improved handling of tool-heavy responses in Nova: large tool outputs are compacted before storage or resend, yielding faster, more reliable answers with lower cost and fewer surprise follow-ups.
Reuse of stable tool definitions and instructions across repeated turns reduces redundant processing for tool-driven workflows.

Extended caching of stable conversation history and prior tool-call results so older context can be reused rather than reprocessed, keeping conversations snappier and preserving continuity across turns.
Optimized cache reads/writes to make repeated interactions more efficient and predictable.

Upgraded Usage Widget with clear, compact breakdowns of input, output, cache reads, and cache writes; includes Normal and Verbose views and a responsive narrow-panel layout with collapsible sections for easier review.
Separately displays approximate tool payload size alongside billable model usage and provides more accurate per-section cost calculations to help teams interpret and control usage costs.

Improved change lookup reliability and a lean search + focused details workflow so Nova can list and surface tracked changes more quickly and precisely.

Introduced a drag-and-drop Kanban for experiments spanning Idea, Running, Analyzing, and Concluded states, helping teams manage experiment lifecycles visually and consistently.
New experiment cards surface key details including owner, hypothesis, parent bet, source, and learning counts to make prioritization and triage faster.
Added a full-featured Experiment Detail modal/drawer for editing methods, expected signals, outcomes, and links to hypotheses and learnings.
Reorganized Discovery navigation around Experiments, Validation Tree, Learnings, and External Sources for clearer workflows.
Built AI-assisted synthesis to produce Draft Learnings and surfaced theme/signal suggestions to accelerate insight capture.

Added preview-driven External Sources Import with connectors to Jira Product Discovery, Productboard, and Aha, allowing teams to review suggested classifications, skip items, or link imported work directly to existing outcome entities.